- Blind mode tutorial
lichess.org
Donate

Which chess engine is best in a particular phase?

#11

Very nice that it has all of that flexibility. But you won't find my suggestion within those options. If you did find, then the hypothesis would certainly already be tested by someone else. It would actually need many cycles of testing and parameter tuning before reaching the hypothetical goal: a hybrid engine that can defeat Stockfish.

Very nice that it has all of that flexibility. But you won't find my suggestion within those options. If you did find, then the hypothesis would certainly already be tested by someone else. It would actually need many cycles of testing and parameter tuning before reaching the hypothetical goal: a hybrid engine that can defeat Stockfish.
#12

The best candidate moves in percentage form would probably be more useful for me than finding the right combo of engines to beat another engine.

My goal is finding the right engine for a particular candidate move. Maybe we need to sort engines by their qualities, to create better choices of candidate moves.

Maybe the solution is in sorting the repertoires of these chess engines, which would give us more chess insights.

It would be great to see different engine "Chess insights" on lichess.

The best candidate moves in percentage form would probably be more useful for me than finding the right combo of engines to beat another engine. My goal is finding the right engine for a particular candidate move. Maybe we need to sort engines by their qualities, to create better choices of candidate moves. Maybe the solution is in sorting the repertoires of these chess engines, which would give us more chess insights. It would be great to see different engine "Chess insights" on lichess.
#13

@Toscani said in #12:

It would be great to see different engine "Chess insights" on lichess.

I do think that it might be brewing in that direction. Lichess is working on the possibility to dock own custom engine, if i understood the bits i read here or on github.

It might take some time, i presume to get it working, the bigger the lichess population is, I guess, the more careful one might have to be before letting loose such a new set of features in the live lichess.

Since engine analysis has many tentacles everywhere in many chess features.. It is likely not a very easy to isolate new set of features.

But, i would think it probable (i was going to say logical, but then i became careful of my words, something rare, or is it?).

that insight would likely be benefiting from that bridging construction.

I concur with @EvilChess, there is some need for some data analysis, shareable and well done enough (?), before, during and after such combination experiments.. Specifying the context parameters of strenght and weakness.. Without external referential to the quality of a move, how does one know which engine move is better? Maybe if one has some well defined contexts, where such thing could be tested, and if that context had some relationship with a more general context, one could find threads or principles along which to combine different partially sub-optimal parts.

I will try to find the reference to the combination of experts article.. (before thread archiving, i promise).

@Toscani said in #12: > It would be great to see different engine "Chess insights" on lichess. I do think that it might be brewing in that direction. Lichess is working on the possibility to dock own custom engine, if i understood the bits i read here or on github. It might take some time, i presume to get it working, the bigger the lichess population is, I guess, the more careful one might have to be before letting loose such a new set of features in the live lichess. Since engine analysis has many tentacles everywhere in many chess features.. It is likely not a very easy to isolate new set of features. But, i would think it probable (i was going to say logical, but then i became careful of my words, something rare, or is it?). that insight would likely be benefiting from that bridging construction. I concur with @EvilChess, there is some need for some data analysis, shareable and well done enough (?), before, during and after such combination experiments.. Specifying the context parameters of strenght and weakness.. Without external referential to the quality of a move, how does one know which engine move is better? Maybe if one has some well defined contexts, where such thing could be tested, and if that context had some relationship with a more general context, one could find threads or principles along which to combine different partially sub-optimal parts. I will try to find the reference to the combination of experts article.. (before thread archiving, i promise).
#14

I wonder about the idea of engine repertoire.

the same experiments where one uses books and engine X engine battles, could be viewed as engine take on openings, or they could be viewed as book (repertoire) take on engines..

They could define the engine biases with respect to human knowledge... Then our faith or loyalties would get tested..(half-kidding).

What should we use?

I suggest that more data points (i.e. sufficiently different engines design) would be better than just one.....
And it seems that things may go in that direction. I am liking this thread.

I wonder about the idea of engine repertoire. the same experiments where one uses books and engine X engine battles, could be viewed as engine take on openings, or they could be viewed as book (repertoire) take on engines.. They could define the engine biases with respect to human knowledge... Then our faith or loyalties would get tested..(half-kidding). What should we use? I suggest that more data points (i.e. sufficiently different engines design) would be better than just one..... And it seems that things may go in that direction. I am liking this thread.
#15

@Toscani said in #10:

Can you give more explanation about how those parameters affect the combination decisions?.
are these assuming some scheme of combination, and those choice of parameter names make sense in such scheme?
e.g. Random, would be there is a pool of parallel accessible engines, and which gets to search from the root, via random picks in that pool, at each new ply.

Or that the tree search part of a single root search is assigned to one engine, but it leaf evaluation function is using that of another engine, if those were made modular and switchable that way (which i would bet, with some prior understanding that they are not, while in chess land one would assume that they should be made so, however, this is nowhere a specification in engine tournament, that some of their meaningful chess module be part of UCI accessible functions.

Some engine may have had the initiative to provide for some partial version that could take a FEN as input, and produce some output, but either the output is incomplete, or does not appear to be the one used with the tree search leafs... I think the driving bottleneck there is at tournament norms.. (whatever the details, that it what serves as specifications for developers to use their CS skills to CS optimize their code, with ELO dependent on those specifications. engine ELO.

I give this impossible scenario as example, of another scheme that might still have used some of those variables but would mean different thing.

I am more curious about something you did not mention. Those features unique to lucas chess. Are they metrics of the chess content, or perhaps via chess they could be metrics of the engine used with lucas chess.

Could you mention the list of those "dimensions" that this GUI is able to provide upon some input (also to be mentioned, many games, engine tournament results as input)..

The rest of this i backed up.. i tend to give too much in one shot.. burying each parts.. if the above does not seem justified.. then i might elaborate. I think i can use NNue innovation to explain. but it needs space...

@Toscani said in #10: > Can you give more explanation about how those parameters affect the combination decisions?. are these assuming some scheme of combination, and those choice of parameter names make sense in such scheme? e.g. Random, would be there is a pool of parallel accessible engines, and which gets to search from the root, via random picks in that pool, at each new ply. Or that the tree search part of a single root search is assigned to one engine, but it leaf evaluation function is using that of another engine, if those were made modular and switchable that way (which i would bet, with some prior understanding that they are not, while in chess land one would assume that they should be made so, however, this is nowhere a specification in engine tournament, that some of their meaningful chess module be part of UCI accessible functions. Some engine may have had the initiative to provide for some partial version that could take a FEN as input, and produce some output, but either the output is incomplete, or does not appear to be the one used with the tree search leafs... I think the driving bottleneck there is at tournament norms.. (whatever the details, that it what serves as specifications for developers to use their CS skills to CS optimize their code, with ELO dependent on those specifications. engine ELO. I give this impossible scenario as example, of another scheme that might still have used some of those variables but would mean different thing. I am more curious about something you did not mention. Those features unique to lucas chess. Are they metrics of the chess content, or perhaps via chess they could be metrics of the engine used with lucas chess. Could you mention the list of those "dimensions" that this GUI is able to provide upon some input (also to be mentioned, many games, engine tournament results as input).. The rest of this i backed up.. i tend to give too much in one shot.. burying each parts.. if the above does not seem justified.. then i might elaborate. I think i can use NNue innovation to explain. but it needs space...
#17

@Toscani , imagine you are going to play an important chess match and you can receive guidance from two GMs. They are named Leela and named Fish. But they won't communicate with each other, just with you. So your best shot is to ask Fish, the stronger GM, if he is convinced about what is the best move at this position, and it he is you just follow his advice, making that move. However, if Fish is not absolutely sure about that being the best move, if he keeps changing his mind as he looks deeper, or if there are two or three moves that tie together as the best, then you would want to check out the opinion of GM Leela.

Then you make the same question to her. If she is absolutely sure of the best move, then she probably sees something that GM Fish doesn't. She is more positional, so may have seen some positional advantage. So you go with her advice. On the other hand, if she is also unsure about the best move, then you filter the moves that have been picked as best moves by GM Fish, and from those you allow GM Leela to pick one.

This is the first strategy I would use to combine two chess engines.

@Toscani , imagine you are going to play an important chess match and you can receive guidance from two GMs. They are named Leela and named Fish. But they won't communicate with each other, just with you. So your best shot is to ask Fish, the stronger GM, if he is convinced about what is the best move at this position, and it he is you just follow his advice, making that move. However, if Fish is not absolutely sure about that being the best move, if he keeps changing his mind as he looks deeper, or if there are two or three moves that tie together as the best, then you would want to check out the opinion of GM Leela. Then you make the same question to her. If she is absolutely sure of the best move, then she probably sees something that GM Fish doesn't. She is more positional, so may have seen some positional advantage. So you go with her advice. On the other hand, if she is also unsure about the best move, then you filter the moves that have been picked as best moves by GM Fish, and from those you allow GM Leela to pick one. This is the first strategy I would use to combine two chess engines.
#18

btw, in my understanding, for a chess engine, the analysis of a candidate move is the same as a regular board position - the position that results from that candidate move.

btw, in my understanding, for a chess engine, the analysis of a candidate move is the same as a regular board position - the position that results from that candidate move.
#19

By noting the enabled Bot lichess insights, we could discover the strengths and weaknesses of the bots and which to use in example a phase or opening, middle or end game lines.
https://lichess.org/player/bots

By gathering up the lichess insights from the bot's, the answer to the subject question could then be answered.

https://youtu.be/LCvQPM4VgkY

https://lichess.org/blog/WvDNticAAMu_mHKP/welcome-lichess-bots
Lichess Bots cannot:
Participate in tournament or simuls
Use lobby seeks or pools
Appear in leaderboards

With the above restrictions it's hard to discover which is best in a particular opening or phase of the game.

Feature Request: Lichess Bot Only Leaderboard to see Tactical awareness vs Opening Family (Strength by phase and ECO code).

By noting the enabled Bot lichess insights, we could discover the strengths and weaknesses of the bots and which to use in example a phase or opening, middle or end game lines. https://lichess.org/player/bots By gathering up the lichess insights from the bot's, the answer to the subject question could then be answered. https://youtu.be/LCvQPM4VgkY https://lichess.org/blog/WvDNticAAMu_mHKP/welcome-lichess-bots Lichess Bots cannot: Participate in tournament or simuls Use lobby seeks or pools Appear in leaderboards With the above restrictions it's hard to discover which is best in a particular opening or phase of the game. Feature Request: Lichess Bot Only Leaderboard to see Tactical awareness vs Opening Family (Strength by phase and ECO code).
#20

I was going to delete my last post.. I think already being able to do combinations, is something.. I was just pushing for having more control or knoweldge about the basis for wanting to combine things.. not just how to combine them. How to chose the parts to combine before combining them.. And having the ability to use position information fully, to determine which engine should be applied. And that means being able to characterize an engine as a function of positon as input.. Its expertise region on the set of all positions.... (minimally diagrams with turn information, castlle and en passant).. but that may be sci-fi.

So that would be complementary to the banksguiA facilty (bad with names here).. i was a bit exagerating in this thread context by bringing the ability for get more information out of engine output than their tounrnament specification forces them to keep in mind. This is not the first thing to do. It might be true, but not necessarily helpful here. Might want to start with combining and see what it gives.. without a priori doubts preventing the experimentation.

edit: u have some reading backkog here too

I was going to delete my last post.. I think already being able to do combinations, is something.. I was just pushing for having more control or knoweldge about the basis for wanting to combine things.. not just how to combine them. How to chose the parts to combine before combining them.. And having the ability to use position information fully, to determine which engine should be applied. And that means being able to characterize an engine as a function of positon as input.. Its expertise region on the set of all positions.... (minimally diagrams with turn information, castlle and en passant).. but that may be sci-fi. So that would be complementary to the banksguiA facilty (bad with names here).. i was a bit exagerating in this thread context by bringing the ability for get more information out of engine output than their tounrnament specification forces them to keep in mind. This is not the first thing to do. It might be true, but not necessarily helpful here. Might want to start with combining and see what it gives.. without a priori doubts preventing the experimentation. edit: u have some reading backkog here too

This topic has been archived and can no longer be replied to.